Nemotron models underpin document AI for real-time business insights
NVIDIA highlights customer deployments using open models for document parsing and analytics
NVIDIA has said its Nemotron open models are helping companies turn unstructured documents into real-time business intelligence by combining retrieval-augmented generation with multi-agent systems.
In a blog post, the chipmaker described how intelligent document processing built on Nemotron can extract structured insights from reports, spreadsheets, web pages and other multimodal files using a combination of extraction models, optical character recognition, vector search, reranking and parsing.
These components form a document intelligence pipeline, packaged as NVIDIA NIM microservices and foundation models that run on NVIDIA graphics processing units. The post said an LLM router can dynamically select between proprietary and open models to match task requirements.
Nemotron Parse, the company’s tool for spatial parsing of tables and text, has been integrated into several commercial workflows.
Chargeback platform Justt.ai uses the system to automate evidence collection and lifecycle management for transaction disputes, with hotel operator HEI Hotels & Resorts among its users. Contract software provider Docusign is evaluating the tool to extract metadata, tables and text from complex PDF files. The company processes millions of transactions daily across more than 1.8 million customers and one billion users.
Related reading
- NVIDIA debuts Nemotron 3 open models for agentic AIs
- Department of Education and Microsoft expand AI partnership in schools
- Meta expands anti-scam drive with fresh campaign featuring Neena Gupta
Edison Scientific has embedded the technology into its PaperQA2 pipeline to help researchers index scientific literature, ground responses in cited passages and improve speed and answer quality.
The post includes tutorials and sample code for teams looking to build custom document pipelines using Nemotron Parse and retrieval-augmented generation. Resources are available through GitHub, Hugging Face, the NGC model catalogue and NVIDIA’s developer platform.
The Recap
- NVIDIA showcased Nemotron-based AI agents for document intelligence.
- Docusign serves more than 1.8 million customers and over one billion users.
- Tutorials and code available on GitHub, Hugging Face and build.nvidia.com.
